/**
* Make a simple fragment texture shader which reads the texture unit 0 and 1
* and writes it as depth and stencil, respectively.
*/
void *
util_make_fragment_tex_shader_writedepthstencil(struct pipe_context *pipe,
unsigned tex_target,
unsigned interp_mode,
bool load_level_zero,
bool use_txf)
{
struct ureg_program *ureg;
struct ureg_src depth_sampler, stencil_sampler;
struct ureg_src tex;
struct ureg_dst out, depth, stencil;
struct ureg_src imm;
ureg = ureg_create( PIPE_SHADER_FRAGMENT );
if (!ureg)
return NULL;
depth_sampler = ureg_DECL_sampler( ureg, 0 );
ureg_DECL_sampler_view(ureg, 0, tex_target,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT);
stencil_sampler = ureg_DECL_sampler( ureg, 1 );
ureg_DECL_sampler_view(ureg, 0, tex_target,
TGSI_RETURN_TYPE_UINT,
TGSI_RETURN_TYPE_UINT,
TGSI_RETURN_TYPE_UINT,
TGSI_RETURN_TYPE_UINT);
tex = ureg_DECL_fs_input( ureg,
TGSI_SEMANTIC_GENERIC, 0,
interp_mode );
out = ureg_DECL_output( ureg,
TGSI_SEMANTIC_COLOR,
0 );
depth = ureg_DECL_output( ureg,
TGSI_SEMANTIC_POSITION,
0 );
stencil = ureg_DECL_output( ureg,
TGSI_SEMANTIC_STENCIL,
0 );
imm = ureg_imm4f( ureg, 0, 0, 0, 1 );
ureg_MOV( ureg, out, imm );
ureg_load_tex(ureg, ureg_writemask(depth, TGSI_WRITEMASK_Z), tex,
depth_sampler, tex_target, load_level_zero, use_txf);
ureg_load_tex(ureg, ureg_writemask(stencil, TGSI_WRITEMASK_Y), tex,
stencil_sampler, tex_target, load_level_zero, use_txf);
ureg_END( ureg );
return ureg_create_shader_and_destroy( ureg, pipe );
}
void *
util_make_fs_msaa_resolve(struct pipe_context *pipe,
enum tgsi_texture_type tgsi_tex, unsigned nr_samples,
enum tgsi_return_type stype)
{
struct ureg_program *ureg;
struct ureg_src sampler, coord;
struct ureg_dst out, tmp_sum, tmp_coord, tmp;
unsigned i;
ureg = ureg_create(PIPE_SHADER_FRAGMENT);
if (!ureg)
return NULL;
/* Declarations. */
sampler = ureg_DECL_sampler(ureg, 0);
ureg_DECL_sampler_view(ureg, 0, tgsi_tex, stype, stype, stype, stype);
coord = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_GENERIC, 0,
TGSI_INTERPOLATE_LINEAR);
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
tmp_sum = ureg_DECL_temporary(ureg);
tmp_coord = ureg_DECL_temporary(ureg);
tmp = ureg_DECL_temporary(ureg);
/* Instructions. */
ureg_MOV(ureg, tmp_sum, ureg_imm1f(ureg, 0));
ureg_F2U(ureg, tmp_coord, coord);
for (i = 0; i < nr_samples; i++) {
/* Read one sample. */
ureg_MOV(ureg, ureg_writemask(tmp_coord, TGSI_WRITEMASK_W),
ureg_imm1u(ureg, i));
ureg_TXF(ureg, tmp, tgsi_tex, ureg_src(tmp_coord), sampler);
if (stype == TGSI_RETURN_TYPE_UINT)
ureg_U2F(ureg, tmp, ureg_src(tmp));
else if (stype == TGSI_RETURN_TYPE_SINT)
ureg_I2F(ureg, tmp, ureg_src(tmp));
/* Add it to the sum.*/
ureg_ADD(ureg, tmp_sum, ureg_src(tmp_sum), ureg_src(tmp));
}
/* Calculate the average and return. */
ureg_MUL(ureg, tmp_sum, ureg_src(tmp_sum),
ureg_imm1f(ureg, 1.0 / nr_samples));
if (stype == TGSI_RETURN_TYPE_UINT)
ureg_F2U(ureg, out, ureg_src(tmp_sum));
else if (stype == TGSI_RETURN_TYPE_SINT)
ureg_F2I(ureg, out, ureg_src(tmp_sum));
else
ureg_MOV(ureg, out, ureg_src(tmp_sum));
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
static void *
create_frag_shader(struct vl_matrix_filter *filter, unsigned num_offsets,
struct vertex2f *offsets, const float *matrix_values)
{
struct ureg_program *shader;
struct ureg_src i_vtex;
struct ureg_src sampler;
struct ureg_dst tmp;
struct ureg_dst t_sum;
struct ureg_dst o_fragment;
unsigned i;
shader = ureg_create(PIPE_SHADER_FRAGMENT);
if (!shader) {
return NULL;
}
i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR);
sampler = ureg_DECL_sampler(shader, 0);
ureg_DECL_sampler_view(shader, 0, TGSI_TEXTURE_2D,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT);
tmp = ureg_DECL_temporary(shader);
t_sum = ureg_DECL_temporary(shader);
o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
ureg_MOV(shader, t_sum, ureg_imm1f(shader, 0.0f));
for (i = 0; i < num_offsets; ++i) {
if (matrix_values[i] == 0.0f)
continue;
if (!is_vec_zero(offsets[i])) {
ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XY),
i_vtex, ureg_imm2f(shader, offsets[i].x, offsets[i].y));
ureg_MOV(shader, ureg_writemask(tmp, TGSI_WRITEMASK_ZW),
ureg_imm1f(shader, 0.0f));
ureg_TEX(shader, tmp, TGSI_TEXTURE_2D, ureg_src(tmp), sampler);
} else {
ureg_TEX(shader, tmp, TGSI_TEXTURE_2D, i_vtex, sampler);
}
ureg_MAD(shader, t_sum, ureg_src(tmp), ureg_imm1f(shader, matrix_values[i]),
ureg_src(t_sum));
}
ureg_MOV(shader, o_fragment, ureg_src(t_sum));
ureg_END(shader);
return ureg_create_shader_and_destroy(shader, filter->pipe);
}
/**
* Called when a new variant is needed, we need to translate
* the ATI fragment shader to TGSI
*/
enum pipe_error
st_translate_atifs_program(
struct ureg_program *ureg,
struct ati_fragment_shader *atifs,
struct gl_program *program,
GLuint numInputs,
const GLuint inputMapping[],
const ubyte inputSemanticName[],
const ubyte inputSemanticIndex[],
const GLuint interpMode[],
GLuint numOutputs,
const GLuint outputMapping[],
const ubyte outputSemanticName[],
const ubyte outputSemanticIndex[])
{
enum pipe_error ret = PIPE_OK;
unsigned pass, i, r;
struct st_translate translate, *t;
t = &translate;
memset(t, 0, sizeof *t);
t->inputMapping = inputMapping;
t->outputMapping = outputMapping;
t->ureg = ureg;
t->atifs = atifs;
/*
* Declare input attributes.
*/
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_fs_input(ureg,
inputSemanticName[i],
inputSemanticIndex[i],
interpMode[i]);
}
/*
* Declare output attributes:
* we always have numOutputs=1 and it's FRAG_RESULT_COLOR
*/
t->outputs[0] = ureg_DECL_output(ureg,
TGSI_SEMANTIC_COLOR,
outputSemanticIndex[0]);
/* Emit constants and immediates. Mesa uses a single index space
* for these, so we put all the translated regs in t->constants.
*/
if (program->Parameters) {
t->constants = calloc(program->Parameters->NumParameters,
sizeof t->constants[0]);
if (t->constants == NULL) {
ret = PIPE_ERROR_OUT_OF_MEMORY;
goto out;
}
for (i = 0; i < program->Parameters->NumParameters; i++) {
switch (program->Parameters->Parameters[i].Type) {
case PROGRAM_STATE_VAR:
case PROGRAM_UNIFORM:
t->constants[i] = ureg_DECL_constant(ureg, i);
break;
case PROGRAM_CONSTANT:
t->constants[i] =
ureg_DECL_immediate(ureg,
(const float*)program->Parameters->ParameterValues[i],
4);
break;
default:
break;
}
}
}
/* texture samplers */
for (i = 0; i < MAX_NUM_FRAGMENT_REGISTERS_ATI; i++) {
if (program->SamplersUsed & (1 << i)) {
t->samplers[i] = ureg_DECL_sampler(ureg, i);
/* the texture target is still unknown, it will be fixed in the draw call */
ureg_DECL_sampler_view(ureg, i, TGSI_TEXTURE_2D,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT);
}
}
/* emit instructions */
for (pass = 0; pass < atifs->NumPasses; pass++) {
t->current_pass = pass;
for (r = 0; r < MAX_NUM_FRAGMENT_REGISTERS_ATI; r++) {
struct atifs_setupinst *texinst = &atifs->SetupInst[pass][r];
compile_setupinst(t, r, texinst);
}
for (i = 0; i < atifs->numArithInstr[pass]; i++) {
struct atifs_instruction *inst = &atifs->Instructions[pass][i];
compile_instruction(t, inst);
}
}
finalize_shader(t, atifs->NumPasses);
out:
free(t->constants);
if (t->error) {
debug_printf("%s: translate error flag set\n", __func__);
}
return ret;
}
void *
util_make_fs_msaa_resolve_bilinear(struct pipe_context *pipe,
enum tgsi_texture_type tgsi_tex,
unsigned nr_samples,
enum tgsi_return_type stype)
{
struct ureg_program *ureg;
struct ureg_src sampler, coord;
struct ureg_dst out, tmp, top, bottom;
struct ureg_dst tmp_coord[4], tmp_sum[4];
unsigned i, c;
ureg = ureg_create(PIPE_SHADER_FRAGMENT);
if (!ureg)
return NULL;
/* Declarations. */
sampler = ureg_DECL_sampler(ureg, 0);
ureg_DECL_sampler_view(ureg, 0, tgsi_tex, stype, stype, stype, stype);
coord = ureg_DECL_fs_input(ureg, TGSI_SEMANTIC_GENERIC, 0,
TGSI_INTERPOLATE_LINEAR);
out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
for (c = 0; c < 4; c++)
tmp_sum[c] = ureg_DECL_temporary(ureg);
for (c = 0; c < 4; c++)
tmp_coord[c] = ureg_DECL_temporary(ureg);
tmp = ureg_DECL_temporary(ureg);
top = ureg_DECL_temporary(ureg);
bottom = ureg_DECL_temporary(ureg);
/* Instructions. */
for (c = 0; c < 4; c++)
ureg_MOV(ureg, tmp_sum[c], ureg_imm1f(ureg, 0));
/* Get 4 texture coordinates for the bilinear filter. */
ureg_F2U(ureg, tmp_coord[0], coord); /* top-left */
ureg_UADD(ureg, tmp_coord[1], ureg_src(tmp_coord[0]),
ureg_imm4u(ureg, 1, 0, 0, 0)); /* top-right */
ureg_UADD(ureg, tmp_coord[2], ureg_src(tmp_coord[0]),
ureg_imm4u(ureg, 0, 1, 0, 0)); /* bottom-left */
ureg_UADD(ureg, tmp_coord[3], ureg_src(tmp_coord[0]),
ureg_imm4u(ureg, 1, 1, 0, 0)); /* bottom-right */
for (i = 0; i < nr_samples; i++) {
for (c = 0; c < 4; c++) {
/* Read one sample. */
ureg_MOV(ureg, ureg_writemask(tmp_coord[c], TGSI_WRITEMASK_W),
ureg_imm1u(ureg, i));
ureg_TXF(ureg, tmp, tgsi_tex, ureg_src(tmp_coord[c]), sampler);
if (stype == TGSI_RETURN_TYPE_UINT)
ureg_U2F(ureg, tmp, ureg_src(tmp));
else if (stype == TGSI_RETURN_TYPE_SINT)
ureg_I2F(ureg, tmp, ureg_src(tmp));
/* Add it to the sum.*/
ureg_ADD(ureg, tmp_sum[c], ureg_src(tmp_sum[c]), ureg_src(tmp));
}
}
/* Calculate the average. */
for (c = 0; c < 4; c++)
ureg_MUL(ureg, tmp_sum[c], ureg_src(tmp_sum[c]),
ureg_imm1f(ureg, 1.0 / nr_samples));
/* Take the 4 average values and apply a standard bilinear filter. */
ureg_FRC(ureg, tmp, coord);
ureg_LRP(ureg, top,
ureg_scalar(ureg_src(tmp), 0),
ureg_src(tmp_sum[1]),
ureg_src(tmp_sum[0]));
ureg_LRP(ureg, bottom,
ureg_scalar(ureg_src(tmp), 0),
ureg_src(tmp_sum[3]),
ureg_src(tmp_sum[2]));
ureg_LRP(ureg, tmp,
ureg_scalar(ureg_src(tmp), 1),
ureg_src(bottom),
ureg_src(top));
/* Convert to the texture format and return. */
if (stype == TGSI_RETURN_TYPE_UINT)
ureg_F2U(ureg, out, ureg_src(tmp));
else if (stype == TGSI_RETURN_TYPE_SINT)
ureg_F2I(ureg, out, ureg_src(tmp));
else
ureg_MOV(ureg, out, ureg_src(tmp));
ureg_END(ureg);
return ureg_create_shader_and_destroy(ureg, pipe);
}
/**
* Make simple fragment texture shader:
* IMM {0,0,0,1} // (if writemask != 0xf)
* MOV TEMP[0], IMM[0] // (if writemask != 0xf)
* TEX TEMP[0].writemask, IN[0], SAMP[0], 2D;
* .. optional SINT <-> UINT clamping ..
* MOV OUT[0], TEMP[0]
* END;
*
* \param tex_target one of PIPE_TEXTURE_x
* \parma interp_mode either TGSI_INTERPOLATE_LINEAR or PERSPECTIVE
* \param writemask mask of TGSI_WRITEMASK_x
*/
void *
util_make_fragment_tex_shader_writemask(struct pipe_context *pipe,
unsigned tex_target,
unsigned interp_mode,
unsigned writemask,
enum tgsi_return_type stype,
enum tgsi_return_type dtype,
bool load_level_zero,
bool use_txf)
{
struct ureg_program *ureg;
struct ureg_src sampler;
struct ureg_src tex;
struct ureg_dst temp;
struct ureg_dst out;
assert((stype == TGSI_RETURN_TYPE_FLOAT) == (dtype == TGSI_RETURN_TYPE_FLOAT));
assert(interp_mode == TGSI_INTERPOLATE_LINEAR ||
interp_mode == TGSI_INTERPOLATE_PERSPECTIVE);
ureg = ureg_create( PIPE_SHADER_FRAGMENT );
if (!ureg)
return NULL;
sampler = ureg_DECL_sampler( ureg, 0 );
ureg_DECL_sampler_view(ureg, 0, tex_target, stype, stype, stype, stype);
tex = ureg_DECL_fs_input( ureg,
TGSI_SEMANTIC_GENERIC, 0,
interp_mode );
out = ureg_DECL_output( ureg,
TGSI_SEMANTIC_COLOR,
0 );
temp = ureg_DECL_temporary(ureg);
if (writemask != TGSI_WRITEMASK_XYZW) {
struct ureg_src imm = ureg_imm4f( ureg, 0, 0, 0, 1 );
ureg_MOV( ureg, out, imm );
}
if (tex_target == TGSI_TEXTURE_BUFFER)
ureg_TXF(ureg,
ureg_writemask(temp, writemask),
tex_target, tex, sampler);
else
ureg_load_tex(ureg, ureg_writemask(temp, writemask), tex, sampler,
tex_target, load_level_zero, use_txf);
if (stype != dtype) {
if (stype == TGSI_RETURN_TYPE_SINT) {
assert(dtype == TGSI_RETURN_TYPE_UINT);
ureg_IMAX(ureg, temp, ureg_src(temp), ureg_imm1i(ureg, 0));
} else {
assert(stype == TGSI_RETURN_TYPE_UINT);
assert(dtype == TGSI_RETURN_TYPE_SINT);
ureg_UMIN(ureg, temp, ureg_src(temp), ureg_imm1u(ureg, (1u << 31) - 1));
}
}
ureg_MOV(ureg, out, ureg_src(temp));
ureg_END( ureg );
return ureg_create_shader_and_destroy( ureg, pipe );
}
/**
* Translate Mesa program to TGSI format.
* \param program the program to translate
* \param numInputs number of input registers used
* \param inputMapping maps Mesa fragment program inputs to TGSI generic
* input indexes
* \param inputSemanticName the TGSI_SEMANTIC flag for each input
* \param inputSemanticIndex the semantic index (ex: which texcoord) for
* each input
* \param interpMode the TGSI_INTERPOLATE_LINEAR/PERSP mode for each input
* \param numOutputs number of output registers used
* \param outputMapping maps Mesa fragment program outputs to TGSI
* generic outputs
* \param outputSemanticName the TGSI_SEMANTIC flag for each output
* \param outputSemanticIndex the semantic index (ex: which texcoord) for
* each output
*
* \return PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY
*/
enum pipe_error
st_translate_mesa_program(
struct gl_context *ctx,
uint procType,
struct ureg_program *ureg,
const struct gl_program *program,
GLuint numInputs,
const GLuint inputMapping[],
const ubyte inputSemanticName[],
const ubyte inputSemanticIndex[],
const GLuint interpMode[],
GLuint numOutputs,
const GLuint outputMapping[],
const ubyte outputSemanticName[],
const ubyte outputSemanticIndex[])
{
struct st_translate translate, *t;
unsigned i;
enum pipe_error ret = PIPE_OK;
assert(numInputs <= ARRAY_SIZE(t->inputs));
assert(numOutputs <= ARRAY_SIZE(t->outputs));
t = &translate;
memset(t, 0, sizeof *t);
t->procType = procType;
t->inputMapping = inputMapping;
t->outputMapping = outputMapping;
t->ureg = ureg;
/*_mesa_print_program(program);*/
/*
* Declare input attributes.
*/
if (procType == PIPE_SHADER_FRAGMENT) {
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_fs_input(ureg,
inputSemanticName[i],
inputSemanticIndex[i],
interpMode[i]);
}
if (program->info.inputs_read & VARYING_BIT_POS) {
/* Must do this after setting up t->inputs, and before
* emitting constant references, below:
*/
emit_wpos(st_context(ctx), t, program, ureg);
}
/*
* Declare output attributes.
*/
for (i = 0; i < numOutputs; i++) {
switch (outputSemanticName[i]) {
case TGSI_SEMANTIC_POSITION:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_POSITION, /* Z / Depth */
outputSemanticIndex[i] );
t->outputs[i] = ureg_writemask( t->outputs[i],
TGSI_WRITEMASK_Z );
break;
case TGSI_SEMANTIC_STENCIL:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_STENCIL, /* Stencil */
outputSemanticIndex[i] );
t->outputs[i] = ureg_writemask( t->outputs[i],
TGSI_WRITEMASK_Y );
break;
case TGSI_SEMANTIC_COLOR:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_COLOR,
outputSemanticIndex[i] );
break;
default:
debug_assert(0);
return 0;
}
}
}
else if (procType == PIPE_SHADER_GEOMETRY) {
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_input(ureg,
inputSemanticName[i],
inputSemanticIndex[i], 0, 1);
}
for (i = 0; i < numOutputs; i++) {
t->outputs[i] = ureg_DECL_output( ureg,
outputSemanticName[i],
outputSemanticIndex[i] );
}
}
else {
assert(procType == PIPE_SHADER_VERTEX);
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_vs_input(ureg, i);
}
for (i = 0; i < numOutputs; i++) {
t->outputs[i] = ureg_DECL_output( ureg,
outputSemanticName[i],
outputSemanticIndex[i] );
if (outputSemanticName[i] == TGSI_SEMANTIC_FOG) {
/* force register to contain a fog coordinate in the form (F, 0, 0, 1). */
ureg_MOV(ureg,
ureg_writemask(t->outputs[i], TGSI_WRITEMASK_YZW),
ureg_imm4f(ureg, 0.0f, 0.0f, 0.0f, 1.0f));
t->outputs[i] = ureg_writemask(t->outputs[i], TGSI_WRITEMASK_X);
}
}
}
/* Declare address register.
*/
if (program->arb.NumAddressRegs > 0) {
debug_assert( program->arb.NumAddressRegs == 1 );
t->address[0] = ureg_DECL_address( ureg );
}
/* Declare misc input registers
*/
{
GLbitfield sysInputs = program->info.system_values_read;
for (i = 0; sysInputs; i++) {
if (sysInputs & (1 << i)) {
unsigned semName = _mesa_sysval_to_semantic(i);
t->systemValues[i] = ureg_DECL_system_value(ureg, semName, 0);
if (semName == TGSI_SEMANTIC_INSTANCEID ||
semName == TGSI_SEMANTIC_VERTEXID) {
/* From Gallium perspective, these system values are always
* integer, and require native integer support. However, if
* native integer is supported on the vertex stage but not the
* pixel stage (e.g, i915g + draw), Mesa will generate IR that
* assumes these system values are floats. To resolve the
* inconsistency, we insert a U2F.
*/
struct st_context *st = st_context(ctx);
struct pipe_screen *pscreen = st->pipe->screen;
assert(procType == PIPE_SHADER_VERTEX);
assert(pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, PIPE_SHADER_CAP_INTEGERS));
(void) pscreen; /* silence non-debug build warnings */
if (!ctx->Const.NativeIntegers) {
struct ureg_dst temp = ureg_DECL_local_temporary(t->ureg);
ureg_U2F( t->ureg, ureg_writemask(temp, TGSI_WRITEMASK_X), t->systemValues[i]);
t->systemValues[i] = ureg_scalar(ureg_src(temp), 0);
}
}
if (procType == PIPE_SHADER_FRAGMENT &&
semName == TGSI_SEMANTIC_POSITION)
emit_wpos(st_context(ctx), t, program, ureg);
sysInputs &= ~(1 << i);
}
}
}
if (program->arb.IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) {
/* If temps are accessed with indirect addressing, declare temporaries
* in sequential order. Else, we declare them on demand elsewhere.
*/
for (i = 0; i < program->arb.NumTemporaries; i++) {
/* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
t->temps[i] = ureg_DECL_temporary( t->ureg );
}
}
/* Emit constants and immediates. Mesa uses a single index space
* for these, so we put all the translated regs in t->constants.
*/
if (program->Parameters) {
t->constants = calloc( program->Parameters->NumParameters,
sizeof t->constants[0] );
if (t->constants == NULL) {
ret = PIPE_ERROR_OUT_OF_MEMORY;
goto out;
}
for (i = 0; i < program->Parameters->NumParameters; i++) {
switch (program->Parameters->Parameters[i].Type) {
case PROGRAM_STATE_VAR:
case PROGRAM_UNIFORM:
t->constants[i] = ureg_DECL_constant( ureg, i );
break;
/* Emit immediates only when there's no indirect addressing of
* the const buffer.
* FIXME: Be smarter and recognize param arrays:
* indirect addressing is only valid within the referenced
* array.
*/
case PROGRAM_CONSTANT:
if (program->arb.IndirectRegisterFiles & PROGRAM_ANY_CONST)
t->constants[i] = ureg_DECL_constant( ureg, i );
else
t->constants[i] =
ureg_DECL_immediate( ureg,
(const float*) program->Parameters->ParameterValues[i],
4 );
break;
default:
break;
}
}
}
/* texture samplers */
for (i = 0; i < ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits; i++) {
if (program->SamplersUsed & (1u << i)) {
unsigned target =
translate_texture_index(program->TexturesUsed[i],
!!(program->ShadowSamplers & (1 << i)));
t->samplers[i] = ureg_DECL_sampler( ureg, i );
ureg_DECL_sampler_view(ureg, i, target,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT);
}
}
/* Emit each instruction in turn:
*/
for (i = 0; i < program->arb.NumInstructions; i++)
compile_instruction(ctx, t, &program->arb.Instructions[i]);
out:
free(t->constants);
return ret;
}
static void *
create_frag_shader(struct vl_median_filter *filter,
struct vertex2f *offsets,
unsigned num_offsets)
{
struct pipe_screen *screen = filter->pipe->screen;
struct ureg_program *shader;
struct ureg_src i_vtex;
struct ureg_src sampler;
struct ureg_dst *t_array = MALLOC(sizeof(struct ureg_dst) * num_offsets);
struct ureg_dst o_fragment;
const unsigned median = num_offsets >> 1;
unsigned i, j;
assert(num_offsets & 1); /* we need an odd number of offsets */
if (!(num_offsets & 1)) { /* yeah, we REALLY need an odd number of offsets!!! */
FREE(t_array);
return NULL;
}
if (num_offsets > screen->get_shader_param(
screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_TEMPS)) {
FREE(t_array);
return NULL;
}
shader = ureg_create(PIPE_SHADER_FRAGMENT);
if (!shader) {
FREE(t_array);
return NULL;
}
i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR);
sampler = ureg_DECL_sampler(shader, 0);
ureg_DECL_sampler_view(shader, 0, TGSI_TEXTURE_2D,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT,
TGSI_RETURN_TYPE_FLOAT);
for (i = 0; i < num_offsets; ++i)
t_array[i] = ureg_DECL_temporary(shader);
o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
/*
* t_array[0..*] = vtex + offset[0..*]
* t_array[0..*] = tex(t_array[0..*], sampler)
* result = partial_bubblesort(t_array)[mid]
*/
for (i = 0; i < num_offsets; ++i) {
if (!is_vec_zero(offsets[i])) {
ureg_ADD(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_XY),
i_vtex, ureg_imm2f(shader, offsets[i].x, offsets[i].y));
ureg_MOV(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_ZW),
ureg_imm1f(shader, 0.0f));
}
}
for (i = 0; i < num_offsets; ++i) {
struct ureg_src src = is_vec_zero(offsets[i]) ? i_vtex : ureg_src(t_array[i]);
ureg_TEX(shader, t_array[i], TGSI_TEXTURE_2D, src, sampler);
}
// TODO: Couldn't this be improved even more?
for (i = 0; i <= median; ++i) {
for (j = 1; j < (num_offsets - i - 1); ++j) {
struct ureg_dst tmp = ureg_DECL_temporary(shader);
ureg_MOV(shader, tmp, ureg_src(t_array[j]));
ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
ureg_MIN(shader, t_array[j - 1], ureg_src(tmp), ureg_src(t_array[j - 1]));
ureg_release_temporary(shader, tmp);
}
if (i == median)
ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
else
ureg_MIN(shader, t_array[j - 1], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
}
ureg_MOV(shader, o_fragment, ureg_src(t_array[median]));
ureg_END(shader);
FREE(t_array);
return ureg_create_shader_and_destroy(shader, filter->pipe);
}
